Basically-Human-4B is a 4 billion parameter language model developed by GreenerPastures, built on the Qwen3 4B architecture. Fine-tuned for immersive and emotionally resonant character interaction, it excels at maintaining persona and crafting believable responses in dynamic storytelling. This model supports ChatML formatting and is optimized for text-based roleplay, NPC simulation, and interactive fiction, offering a compact yet capable solution for realistic dialogue.
Loading preview...
Basically-Human-4B: Roleplaying Language Model
Basically-Human-4B is a 4 billion parameter language model, based on the Qwen3 4B architecture, specifically fine-tuned for immersive roleplaying and character interaction. It is designed to emulate the nuances of human dialogue, focusing on maintaining persona, generating believable responses, and facilitating dynamic storytelling.
Key Capabilities
- Persona Consistency: Excels at staying in character throughout extended interactions.
- Realistic Dialogue: Generates emotionally resonant and human-like responses.
- Dynamic Storytelling: Capable of driving narrative progression in interactive scenarios.
- ChatML Support: Utilizes the ChatML instruction format with the Qwen3 chat template for structured, multi-turn conversations.
Training and Alignment
The model underwent supervised fine-tuning on a diverse collection of instruct and roleplaying datasets, followed by a Kahneman-Tversky optimization step for final alignment. The training data includes various instruction and roleplaying sets from sources like PocketDoc, NewEden, Epiculous, and Nitral-AI.
Good For
- Text-based Roleplay: Ideal for creating engaging and consistent roleplaying experiences.
- NPC Simulation: Suitable for developing non-player characters with realistic dialogue and behavior.
- Interactive Fiction: Enhances interactive stories with dynamic and responsive character interactions.